YouTube Transcript Scraper

The Actor returns the transcript of any public YouTube video or Short — as timed segments, and as one clean block of plain text ready to summarise, search, or feed to a model. It can also attach the captions as a .vtt or .srt subtitle file. Videos are accepted as URLs or bare IDs, and no YouTube account, API key, or cookie is required.

Accepted input

Field	Type	Default	Description
video_sources	array	—	Required. Video or Shorts URLs, or bare 11-character IDs.
transcript_language	string	—	Language code to request, such as en, de, ja. Leave empty for the video's own default.
includeSubtitleFile	boolean	false	Also attach the captions as a subtitle file.
caption_format	enum	vtt	vtt for WebVTT, srt for SubRip. Applies to the subtitle file only.
maxItems	integer	100	Maximum videos processed. 0 processes every supplied target.

{
  "video_sources": [
    { "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ" },
    { "url": "https://www.youtube.com/shorts/P4-3XTiiogo" },
    { "url": "9bZkp7q19f0" }
  ],
  "maxItems": 100
}

Response fields

One row per video.

Field	Contents
recordType	Always video_transcript
source	The URL or ID as supplied
videoId	Video ID
url	Canonical video URL
text	The whole transcript as plain text, segments joined in order
transcript	Timed segments, each with startMs, endMs, startTime, text
segmentCount	Number of timed segments
wordCount	Approximate word count of text
transcriptLanguage	Language code YouTube actually served
transcriptLanguageName	Display name of that track
position	One-based position within the run

Present only with includeSubtitleFile:

Field	Contents
subtitleText	The captions as a complete .vtt or .srt file
subtitleFormat	vtt or srt
subtitleLanguage	Language code of the attached track
availableLanguages	Every caption track YouTube lists for the video

{
  "recordType": "video_transcript",
  "videoId": "dQw4w9WgXcQ",
  "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "transcriptLanguage": "en",
  "transcriptLanguageName": "English (auto-generated)",
  "segmentCount": 52,
  "wordCount": 291,
  "text": "[Music] We're no strangers to love. You know the rules and so do I...",
  "transcript": [
    { "startMs": "320", "endMs": "14580", "startTime": "0:00", "text": "[Music]" }
  ]
}

Plain text and timed segments

Both are always returned, because they answer different questions.

text is the transcript as a single string with the timing removed. This is what you want for summarising, keyword search, sentiment work, or anything fed to a language model — no post-processing, no joining segments yourself.

transcript keeps YouTube's segment boundaries and timings, which is what you need to jump to a moment in the video, build a clickable transcript, or align quotes with timestamps.

Segments are joined with a space rather than a newline. YouTube breaks auto-generated captions at roughly even intervals rather than at sentence ends, so newlines would imply sentence boundaries that are not really there.

Choosing a language

Leave transcript_language empty and you get the video's own default track, which is the right choice most of the time.

If you do set it, read transcriptLanguage on the row. YouTube does not refuse a language it does not have — it answers successfully with whatever track it holds. Asking a video that carries only English, German, and Japanese for Spanish returns German, with no error. The Actor cannot prevent that, so it does two things instead: every row records the language actually served, and the run log warns when it differs from what you asked for.

"dQw4w9WgXcQ": no es transcript exists, so YouTube returned de-DE instead.
The row records the language actually served.

Filter on transcriptLanguage if a run must contain one language only. With includeSubtitleFile enabled, availableLanguages lists every track the video actually has, which is the reliable way to find out what is on offer.

Subtitle files

includeSubtitleFile attaches the captions as a complete subtitle file in subtitleText — WebVTT by default, SubRip with caption_format: "srt". Both are ready to save and load into a video player or editor as-is.

It costs two extra requests per video: one to list the tracks, one to fetch the chosen one. It is charged once, only when both succeed. The track chosen is the one matching the transcript's own language where possible, so the two halves of a row agree; otherwise the video's first listed track is used.

Translated subtitles are not offered. YouTube advertises over 120 translation targets for a typical video, but every request for a translated track fails at the source, so the option is left out rather than shipped as a switch that never works.

Behaviour on partial results

Each video is fetched individually. A target that cannot be parsed, that has no transcript, or that returns an empty one is logged and skipped, and the run continues with the rest. The run fails only when no supplied target was a usable video reference.

Not every video has a transcript. Captions must exist — either uploaded by the creator or auto-generated by YouTube — and creators can disable them. Music videos, very short clips, and videos in languages YouTube does not auto-caption often have none. A skipped video costs nothing.

Frequently asked questions

Which URL forms are accepted? youtube.com/watch?v=, youtu.be/, youtube.com/shorts/, /embed/, and /live/, plus bare 11-character video IDs. Mixing forms in one run is fine.

Does this work on Shorts? Yes. Shorts use the same caption system as regular videos and need no special handling — pass the /shorts/ URL or the bare ID.

Why did a video return no transcript? It has no caption track. Auto-generated captions cover many but not all languages, and creators can turn captions off entirely. The run log names each skipped video, and skipped videos are not charged.

Are auto-generated captions distinguishable from human ones? Yes, from transcriptLanguageName: YouTube labels automatic tracks explicitly, as in English (auto-generated). Human-written tracks carry the plain language name.

Is the transcript punctuated? For auto-generated tracks, partially — YouTube adds sentence punctuation but no speaker labels, and its accuracy varies with audio quality and accent. Human-uploaded tracks are as the creator wrote them.

Can I get a transcript in a language the video was not captioned in? No. Only tracks YouTube actually holds for that video can be returned, and translation of caption tracks fails at the source. Translate the returned text yourself if you need another language.

Is a YouTube account or API key required? No. Only publicly visible captions are read, with no credentials or logged-in session, and no YouTube Data API quota is consumed.

Related Actors

Actor	Purpose
YouTube Video Scraper	Full video metadata, with transcripts as an optional extra
YouTube Channel Scraper	Find every video a channel has published
YouTube Search Scraper	Find videos by keyword, hashtag, or chart
YouTube Playlist Scraper	Every video in a playlist, with optional transcripts